Agility Robotics
Robotics Software Engineer III, Manipulation
About this role
Join Agility's Skills team to develop manipulation behaviors for commercially deployed humanoid robots. You'll own task verticals end-to-end—from data collection and policy training to real-world deployment—using both learned and classical approaches to enable robots operating in warehouses and manufacturing.
What you'll do
- Own task verticals from definition through reliable on-robot execution
- Run iterative learning-from-demonstration flywheel: data collection, annotation, training, evaluation, and deployment
- Perform root-cause failure analysis using logs, replays, and state distribution analysis
- Develop grading rubrics and evaluation workflows to measure policy reliability on real hardware
- Integrate learned policies with higher-level skills for complete robot workflows
- Partner with ML team on base behavior models and task-specific policy post-training
What they're looking for
- Python and software engineering fundamentals
- Robot manipulation and grasping (grasp planning, force control, visual servoing)
- Learning from demonstration and imitation learning
- Reinforcement learning
- Robot data collection and hardware testing
- Robot perception and object recognition
- Teleoperation and sensor integration
- Debugging and log analysis
Opens the official application on the employer’s site. No login required.
Agility Robotics
Agility Robotics builds next-generation humanoid robots designed for commercial deployment in warehouses and manufacturing environments. The company is hiring thermal and mechanical engineers, data infrastructure specialists, robotics skills developers, and pre-sales solutions engineers to support robot design, performance validation, fleet operations, and enterprise customer adoption.
- Website
- agilityrobotics.com
Likely interview questions
- Describe a manipulation task you've deployed on a real robot—what were the biggest failure modes and how did you diagnose them?
- Walk us through your approach to designing a data collection and annotation pipeline for a new manipulation behavior.